If you have lots of dropout, then normalising by matching to a template
without dropout is likely to result in poorer alignment of homologous
structures.
If you have distortions between fMRI and anatomical images, then
rigid-body coregistration will be a poor model, so spatial normalisation
via the anatomical won't work so well.
There are likely to be different fixes for different data, and finding
the best model will be a matter of empirical exploration. If you have
field maps, then distortion correction should be possible. This should
result in a better coregistration, and therefore more accurate spatial
normalisation via the anatomical image.
Best regards,
-John
On Tue, 2009-11-03 at 18:40 +0100, Guillaume Sescousse wrote:
> Dear SPMers,
> I would like to follow up on a discussion about coregistration started
> here:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=ind0901&L=SPM&P=R42697&m=21075.
>
> I have the same problem as Haakon, i.e. mean EPI badly coregistered to
> T1 in the Z direction for 3 out of 14 subjects, even when using phase
> maps in SPM8 unwarping.
> I noticed that these problematic subjects seemed to be the ones with the
> most susceptibility artefacts, so I used Bas's wise advice and
> normalized my mean unwarped EPI the old fashioned way, i.e. on the EPI
> template. I felt happy at first, as it solved my coregistration problem
> for these subjects...
> However I also noticed a strange phenomenon: using the old fashioned
> normalization seemed to reduce my artefacts in the OFC... which can't be
> ! In fact I have the impression that in distorted EPIs this
> normalization process stretches the images to match the undistorted EPI
> template, and hence "creates" signal where there isn't (because of the
> artefacts). I attach an image showing the two normalization processes
> ("new way" with coregistration of EPI on T1 and "old fashioned way"
> using the EPI template). I think that this stretching of the EPI is
> particularly evident at the bottom of the lateral ventricle, which is
> not well coregistered with the T1 scan when using the old fashioned way.
> So I guess my question is: is it really a good idea to use this old
> fashioned normalization process when observing coregistration problems
> likely due to suceptibility artefacts ?
> I apologize for the long e-mail, and will greatly appreciate any comment...
> Thank you very much
>
> Guillaume
>
--
John Ashburner <[log in to unmask]>
|